A 4½-yr monthly stratospheric temperature record derived from TIROS-N Operational Vertical Sounder satellite observations has been used to study the global variability of the stratosphere in the Tropics. A comparison with an independent set of temperatures (Free University of Berlin) is first discussed. Among the different parameters that influence the tropical stratosphere, 1) the regular seasonal cycle, 2) the quasi-biennal oscillation (QBO), and 3) the El Niño–Southern Oscillation (ENSO) effects are studied in detail. A transition level has been found at about 30 hPa. Below this level, the standard stratospheric seasonal cycle in the temperatures is modulated by ENSO and the QBO, while above, ENSO has no discernible influence. In addition, longitudinal variations of monthly mean temperatures show minima during northern winter months from the tropopause up to 50 hPa over some areas, in relation to convection. Results presented here are also discussed in the view of recently published studies based on either radiosonde reports or microwave satellite measurements. While there is a fair agreement with radiosonde-based studies, more finescale details on the horizontal are obtained due to a much better sampling. Differences with other satellite-based studies are due to a better description of the temperature behavior along the vertical.
The climatology of the stratosphere has been a subject of considerable interest since routine radiosonde measurements began in the 1950s. The discovery of the quasi-biennal oscillation (QBO) in tropical stratospheric winds and temperatures by Reed at al. (1961) and Veryard and Ebdon (1961), as well as the first observation in 1952 of a sudden stratospheric warming over northern high latitudes, has motivated many later studies. Nowadays, stratospheric temperature monitoring has become increasingly important mostly for correlative studies of ozone destruction or CO2 increase but also for assessing the impact of volcanic eruptions on climate and for better understanding radiative and dynamic processes that occur in this part of the atmosphere. Only with the relatively recent use of satellites have truly global stratospheric temperature analyses been possible, and in particular the National Oceanic and Atmospheric Administration’s (NOAA) TIROS-N series of sun-synchronous polar-orbiting satellites launched in late 1978. Before satellites, radiosondes provided daily or twice daily temperature measurements at specific points that allowed studies over longer periods of time than possible with the satellite record. The main shortcomings of these data are however that they are not uniformly distributed around the globe, and that changes in the radiosonde systems have occurred, which may lead to problems in long-term monitoring of the temperatures. Raw microwave satellite data also obtained aboard NOAA satellites have been used in the past to infer characteristics of the temperature in the stratosphere (Spencer et al. 1990; Spencer and Christy 1993; Randel and Cobb 1994; Yulaeva and Wallace 1994). While the sensors were shown to be very stable in their calibration, with small intersatellite changes (Spencer and Christy 1993), the problem with the use of these raw data arises from the fact that the part of the atmosphere sampled is rather large (from 150 to 50 hPa). In the Tropics, the problem is even worse since the tropopause occurs near the weighting function maximum (90 hPa), which means that the measured signal is weighted between the upper troposphere and the lower stratosphere. In addition, only one piece of information is available along the vertical, which is another limitation. In this study, we propose an alternative approach based on the use of the temperature profiles that have been retrieved from NOAA satellite radiances. This project is part of a larger one, the NOAA–National Aeronautics and Space Administration Pathfinder program, designed to produce quality satellite-derived data, in particular from the TIROS-N Operational Vertical Sounder (TOVS), which flies aboard NOAA satellites and allows for the determination of temperature profiles in the atmosphere up to 10 hPa. The 3I algorithm is one of the methodologies or paths identified for inverting the TOVS data [referred to as path B, see Susskind et al. (1997) for path A]. Within the frame of this program, we have at the present time at the Laboratoire de Météorologie Dynamique an archive describing the vertical structure of the atmosphere from the surface up to 10 hPa for 53 months (April 1987–August 1991). This record is large enough to address a number of climatological issues such as the temperature variations in the tropical stratosphere. These variations have received much attention since the pioneering study of the annual and semiannual temperature cycles by Reed (1962). It is now well known that at least five distinct forces contribute to the temperature variations in the tropical lower and middle stratosphere:the regular seasonal cycle, the QBO, the semi-annual oscillation (SAO) at higher levels, ENSO, and the radiative effects like those caused by volcanic eruptions (Reid 1994). However, depending on the approach that is adopted (i.e., analysis of radiososonde reports, or of raw satellite data), results may differ, especially regarding the effects of ENSO events on the stratospheric temperatures. Since the present record ends in August 1991, that is, shortly after the Pinatubo eruption, the fifth aspect is not discussed here. The other contributions, with the exception of SAO, which affects levels higher than 10 hPa, will be addressed in this study.
The characteristics of the record used in this study are presented in section 2. The comparison between our record and the dataset of the Stratospheric Group of the Free University of Berlin obtained in a totally independent way is presented in the next section. Then, in section 4, the global variability of the temperatures in the Tropics between 100 and 10 hPa is discussed. More specifically, the seasonal cycle, the role of the QBO, the ENSO effects, and the zonal variations are successively addressed. Finally, in section 5, a summary and a discussion of our results are given.
2. Constitution of a record of stratospheric temperatures and associatedtropospheric variables
TOVS, which flies aboard NOAA polar satellites, consists of three passive vertical sounding instruments (Smith et al., 1979): the High-resolution Infrared Radiation Sounder (HIRS-2), a radiometer with 19 channels in the infrared band and one in the visible band; the Microwave Sounding Unit (MSU), a microwave sounder with 4 channels in the vicinity of 55 GHz; and the Stratospheric Sounding Unit (SSU), a pressure-modulated infrared radiometer with 3 channels near 15 μm. Only HIRS-2 and MSU data have been processed with the 3I method (Chedin et al. 1985) to obtain temperature profiles up to 10 hPa as well as tropospheric water vapor and cloud information. SSU channel 1, with a weighting function maximum at 15 hPa, might have been of interest for our study. However, due to its low spatial resolution (compared to HIRS-2 channels) and coverage, and also due to its rather broad weighting function, it is not used here. The 3I horizontal resolution of 100 by 100 km2 (3I “boxes”) represents a compromise between the spatial resolutions of HIRS-2 (17-km nadir resolution) and MSU (100-km nadir resolution).
At the present time, for the period ranging from April 1987 until August 1991 (i.e., 53 months), the mean temperatures for the following layers, 100–70, 70–50, 50–30, and 30–10 hPa, are available for each day (morning and afternoon separately), each pentad, and each month. In this study, only monthly mean values are used and discussed. Retrievals have been gridded on a 1° lat × 1° long grid. For pentad and monthly means, the gridded values include, in addition to the mean value, the standard deviation and the number of observations for each grid point. Since quality control tests are incorporated into the 3I procedure in order to detect and to reject either bad satellite data or bad retrievals, it may happen that grid points have no observations associated. However, it has been checked that the number of such occurrences was very low, as can be seen in Fig. 1. The number of retrievals, which should optimally be 21 960, lies systematically in the range 21 944–21 960.
All the data used in this study have been retrieved from NOAA-10 observations, which ensures that there is an instrumental consistency and no artificial drift that could be interpreted as a climatic signal. However, to take into account both the possible changes in instruments over their lifetime and possible bias in the simulation of the brightness temperatures necessary for inverting the TOVS data, adjustment coefficients have been calculated and are used in the 3I algorithm. Their determination, based on time and space collocations between TOVS measurements and radiosonde reports makes use of a moving average on a basis of 3 months (R. Armante 1997, personal communication).
Note that, in addition to the above-described products, tropospheric products are also available at the same temporal and spatial resolutions. They include mean temperatures (1000–850-, 850–700-, 700–500-, 500–300-, and 300–100-hPa layers), tropospheric integrated water vapor content, as well as cloud parameters (cloud-top pressure and temperature, cloud fraction for different layers).
The validation of HIRS-2–MSU-derived stratospheric temperatures has been performed in connection with field experiments devoted to the stratosphere therefore, for extremely well-documented events; this ensured the reliability of the stratospheric temperatures measured by the radiosondes. These studies have demonstrated the sensitivity of the retrieved temperatures to low-amplitude thermal anomalies. They have also shown that the retrieved thermal fields reveal to a large extent atmospheric patterns on bidaily and longer timescales for regional and larger space scales (Claud et al. 1993, 1996, 1998).
3. Comparison between TOVS and Berlin stratospheric temperatures
For further validation, monthly mean temperatures retrieved by 3I have been compared to an independent record of stratospheric temperatures obtained at the Free University (FU) of Berlin. The FU Berlin monthly mean temperatures are available for the Northern Hemisphere. They are deduced from daily analyzed temperatures for 0000 UTC, which result from a subjective analysis, using over land the radiosonde observations and over sea the routinely transmitted SATEMs (i.e., thicknesses derived from SSU) and assuring a backward time consistency (e.g., Pawson and Naujokat 1997). Since Berlin data are for levels and not layers, mean temperatures have been deduced, assuming that the temperature within a layer varies linearly between two given levels. In addition, since 10-hPa temperatures from FU Berlin are only available from September until March, comparisons for the 30–10-hPa layer are restricted to these winter months. Finally, comparisons have been performed on a grid of 5° lat × 5° long, which corresponds to the spatial resolution of Berlin data; therefore, the TOVS–3I temperatures have been interpolated to this grid. Whereas comparisons were made for the whole Northern Hemisphere (all latitudes), we present and discuss here only the results obtained over the latitude band spreading from the equator to the latitude of 30°N; the statistics (bias and standard deviation) for the layers 100–50, 50–30, and 30–10 hPa are shown in Fig. 2.
The 100–50-hPa layer is characterized by a systematic negative bias (TOVS–3I < Berlin) of about −0.8 K, while the standard deviation is 0.7 K (Figs. 2a,b). The negative bias is probably due to a mislocation of the tropical tropopause, since it is not present for extratropical regions (statistics not shown).
In the 50–30-hPa layer (Figs. 2c,d), the bias between the two products exhibits a kind of seasonal cycle with TOVS–3I temperatures generally more extreme than Berlin ones (colder during winter, warmer during summer); however, both the bias and the standard deviation remain low, with the exception of 1987/88 winter months, where the latter increases to values larger than 1.7 K.
Finally, comparisons for the 30–10-hPa layer (Figs. 2d,e) exhibit a mean bias also less or of the order of 1 K, while the standard deviation, which is on average 0.8 K, becomes larger in October and November 1990 (1.7 and 2 K, respectively).
Taking into account the time differences, the distinct spatial resolutions, and the assumptions made concerning the vertical variation of the temperature between two levels, the agreement found between the two records is remarkable and gives confidence in the climatic studies that have been conducted and that are discussed below.
4. Characterization of the variability of stratospheric temperatures
A good illustration of the variability of the low stratosphere in the tropical belt (30°S–30°N) is provided by the Hovmöller diagram for the 70–50-hPa layer using meridian mean values for the period April 1987–August 1991, displayed in Fig. 3.
The seasonal cycle appears clearly with minimum temperatures corresponding to northern winter months. In addition to the interannual variations, zonal variations can be seen, like for example the area between about 120°E and 180° being colder during northern winter months than the remaining tropical belt. All these points are investigated in more detail below.
a. The seasonal cycle
Figures 4a–d display the seasonal variations of monthly mean temperatures averaged over the area 30°S–30°N from April 1987 until August 1991 for the 100–70-, 70–50-, 50–30-, and 30–10-hPa layers, respectively. Also represented is the standard deviation (±1σ) around these mean values. The characteristic features of the tropical lower (below 30 hPa) stratospheric temperatures are clearly evident: temperatures are maximum during northern summer months and minimum during northern winter months. There are several factors that may explain the annual cycle (control of extratropical wave forcing, annual variation in the Tropics themselves) but the discrimination between them is still unclear at the present time and is the subject of much debate. The standard deviation around the mean values is maximum in the 100–70-hPa layer, that is, close to the tropopause, where it is about 4 K, and decreases with altitude (about 1.5 K in the 50–30-hPa layer).
Above 30 hPa, two signals can be distinguished, the first, annual, and the second, semiannual. This semiannual temperature variation results from a secondary meridional circulation, which maintains geostrophic balance in the semiannually varying winds at higher levels (e.g., Hirota 1980). Temperatures are maximum in April and in October–November, and minimum in January–February and August. The second maximum peak in the year (October–November), compared to the first one, is of less magnitude, in agreement with earlier studies (Delisi and Dunkerton 1988; Garcia et al. 1997). The 1989 second maximum in October, which corresponds to a phase change of the QBO (from eastern to western), is hardly visible. The standard deviation around the mean values is about 1.5 K.
These results are in excellent agreement with the results of Reid (1994), who obtained the same conclusions using 17 yr of radiosonde temperature measurements for 12 tropical stations.
b. The quasi-biennal oscillation (QBO)
An example of the influence of the QBO is illustrated by the interannual difference between July mean temperatures for 1988 (eastern phase of the QBO) minus 1987 (western phase of the QBO) in the 30–10-hPa layer (Fig. 5). Negative values characterize the equatorial region, in agreement with the fact that colder temperatures are expected during eastern phases of the QBO. The difference lies between 1 and 3.5 K over the region within 15° of the equator, that is, a bit less than the magnitude of the annual cycle shown in Fig. 4. These values are in agreement with the findings of Dunkerton and Delisi (1985). Globally, differences are lower in the Pacific Ocean. The effect of the QBO on lower layers is more difficult to assess since, first, it is relatively less important (Holton and Tan 1980) and, second, there are other driving factors as can be seen below.
c. The ENSO effects
Reid (1994), using the same 17 yr of radiosonde reports from 12 tropical stations showed that the ENSO events cause an additional variation in tropical lower-stratospheric temperatures, namely below about 35 hPa, with the main effect occurring in northern winter months. Spencer et al. (1990) and Yulaeva and Wallace (1994), using channel 4 of MSU, concluded that the ENSO cycle did not show up clearly in the time series of the mean temperature. In contrast, Randel and Cobb (1994), also using channel 4 of the MSU, found that during the northern winter season, the ENSO events create a pair of regions of strong cooling centered at about ± 15° lat in the central Pacific; the cooling decreases toward the equator in both hemispheres, so that these authors point out that the temperature structure mirrors, with opposite sign, the pattern observed in the troposphere. Their analysis was based on the use of 11 yr of MSU observations (from 1982 until 1991) on a grid 5° lat × 10° long.
The picture given by the retrieved stratospheric temperatures for a single ENSO event (1987) has two main characteristics: it is seasonal and the observed patterns are dependent on the layer that is considered. For analyzing the ENSO effects, interannual differences between monthly mean temperatures in 1987 and 1989 (considered here as a “reference year”) will be used; a larger record would be required for considering anomalies. In the 100–70-hPa layer, that is, immediately above the tropical tropopause, the difference between April mean temperatures in 1987 and 1989 indicates a significant cooling over the central Pacific with differences larger than 3 K (Fig. 6). The amplitude of the cooling diminishes with height until 30 hPa, and the 30–10-hPa layer is not affected. Over the western Pacific, only the 100–70-hPa layer is concerned by a cooling in the range 1–2 K. In May, the amplitude of the difference (not shown) has decreased in the 100–70-hPa layer, but has remained constant in the 70–50-hPa layer, whereas the area concerned with the cooling is the same as the previous month. In June (not shown), the difference for the 100–70-hPa layer is nowhere larger than 3 K; in the layer above, two areas with such differences are observed between 10° and 20° (N and S). The same difference but for July and August (Fig. 7) shows regions characterized by negative values over the Pacific between 10° and 30° (N and S), but, close to the equator, values, though small, are positive.
As a conclusion, ENSO events amplify the regular annual cycle over the Pacific by decreasing the temperatures during northern winter months, while the impact can be neglected during the summer season. The cooling is more striking in the central Pacific. It decreases with height from the tropopause up to 30 hPa; above, no effect is discernible.
The differences between our results and those of Randel and Cobb (variations along the vertical; stronger negative response over the eastern equatorial Pacific in our results) are likely to be due to the fact that channel 4 of the MSU peaks at about 90 hPa, while its response function covers the 150–50-hPa pressure range; it is thus indicative not only of the temperature of the low stratosphere but also and to a rather large extent in the Tropics of the temperature of the upper troposphere, as it was already pointed out in Reid (1994).
d. Longitudinal variations
Frederick and Douglass (1983), analyzing temperature measurements obtained over an 8-yr period in the vicinity of the low-latitude tropopause, showed the existence of longitude regions that are consistently colder by approximately 2–3 K than elsewhere in the Tropics. They found that these temperature differences are confined to a layer of thickness 3–5 km centered on the tropopause. Colder regions were found in the longitude band 140°–179°E and up to 70 hPa. The existence of these colder temperatures is essential regarding stratospheric water vapor.
The examination of the variations of monthly mean temperatures over the area 120°E–180° (Fig. 8) and 180°–120°E (Fig. 9) for the four layers reveals that except for the 30–10-hPa and 50–30-hPa layers for which the differences between the two areas are not significant, mean temperatures are colder over the area 120°E–180° compared to the rest of the latitude belt. The differences are more important during northern winter months and decrease with altitude. In the 100–70-hPa layer, during northern winter months, the differences reach 2 K. In contrast to the findings of Frederick and Douglass, differences are also seen above 70 hPa, although smaller than below. In the 70–50-hPa layer, the difference is of about 1 K. In addition, the observation of Fig. 3 and of the same graph, but for the 100–70-hPa layer, both qualitatively very similar, indicates interannual variations of the areas characterized by lower temperatures. On some years, it extends to the west and/or the east (1987/88, 1988/89, 1990/91), while on some others (1989/90) it is more restricted both locally and temporally and minimum temperatures are warmer. A second area, between 40° and 90°W, and a third one, close to 40°E, also exhibit lower temperatures during winter months. Compared to the previous region, the spatial extent is less and the temperature amplitude, though significant, is also smaller (in the 70–50-hPa layer, minima of 202–203 K instead of 201 K in the area 120°E–180°). While these results are globally in agreement with previous observations of, for example, high and cold clouds in the Tropics, typical of major convective centers (e.g., Zhang 1993; Machado and Rossow 1993) and of outgoing longwave radiation (OLR) fields (e.g., Lau and Chan 1983), the data analyzed in this study have the advantage of providing, in addition to the horizontal extent, the vertical extent of the convective signal.
5. Summary and future work
Monthly mean temperature profiles in the stratosphere, as derived from NOAA polar-orbiting satellites through the 3I algorithm, provide information of thermal structures in the Tropics with a good precision. The comparisons performed with an independent set of stratospheric temperatures (Free University of Berlin) show a remarkable agreement, taking into account the time–space differences. In addition, we have checked that there was no big gap in the data (few rejections), which makes them reliable. This good quality of the data has enabled climatological studies. In a first step, the seasonal variability of the temperatures in the stratosphere has been investigated. As expected, below 30 hPa, temperature maxima/minima were found during northern summer/winter months, with a larger deviation around the mean values in the layer just above the tropopause. Above 30 hPa, a semiannual signal is superimposed to the annual one. Our results confirm previous local observations of this cycle, with the second maximum peak in the year (which generally occurs in October–November) with less magnitude than the previous one. In 1989, it is even hardly visible.
This fairly distinct transition at about 30 hPa has also been found in the interannual variations. Below this level, both the QBO and the ENSO cycle modulate the temperature, while above, ENSO does not play any role. Our results, based on one single ENSO event (1987), show that ENSO tends to reinforce the regular seasonal cycle by decreasing the temperatures during Northern Hemisphere winter months. The cooling, of rather low amplitude in the western Pacific (1–2 K for the 100–70-hPa layer), is more striking over the central Pacific (more than 3 K for the same layer) and can be seen in this area up to 50 hPa. No clear effect of ENSO has been observed during summer months. The effect of the QBO has been clearly observed in the 30–10-hPa layer but is more difficult to assess below.
Globally, these results are in fair agreement with results obtained by Reid (1994), using radiosonde reports of 12 tropical stations over a period of 17 yr. However the much better spatial sampling leads to the obtention of more finescale details on the horizontal. Compared to studies based on the use of MSU data, our approach permits us to assess which part of the stratosphere is really affected by ENSO and also gives a more direct inference of the amplitude of this effect.
Concerning the negative response of the lower stratosphere to ENSO during winter months, a study is presently being carried out to assess whether the cooling could be a radiative effect of the optically thick cirrus fields associated with the enhanced convective activity, as suggested by Webster and Stephens (1980). This is made possible by the fact that our record contains consistent cloud information (see section 2).
Finally, longitude regions characterized by colder temperatures during northern winter months have been observed. The spatial extent of the first one varies from year to year around 120°E–180°, while the others are situated between 40° and 90°W and around 40°E, that is, regions known to be convective. In contrast to the local observations, we find that the temperature differences are not confined below 70 hPa but can also be seen up to 50 hPa. The potential of our data to provide both the horizontal and vertical extent of the convective signal will be exploited on a longer temporal record.
The studies described here will be continued using a larger temporal data bank: from 53 months, the data bank will be extended to the whole period for which TOVS data are available, that is, since 1979. In such a context, a more full characterization of the natural variability of the stratospheric temperatures up to 10 hPa will be possible. Phenomena that have not been examined so far (like the effect of volcanic eruptions) will be examined in detail. The determination of temperature trend will also be considered. Besides, as was outlined in section 2, a specificity of this data bank is that it includes not only stratospheric temperatures, but also consistent tropospheric variables. We will thus look at the possible links between the stratosphere and the troposphere.
We are grateful to Prof. K. Labitzke and B. Naujokat (Stratospheric Group, Free University of Berlin) for providing analyzed stratospheric temperatures. Many thanks to the members of the ARA group who have processed the TOVS data, and more particularly to Drs. R. Armante and J. P. Chaboureau, and to L. Crepeau, D. Kasysavanh, B. Bonnet, P. Brockmann, P. Bouster, and S. Dardaillon. CC is indebted to J. Ovarlez for valuable comments on the paper. This work was performed under CEE Contracts EV5V-CT94-0441 and ENV4-CT96-0325. The 3I processing was performed on the computers of the Institut de Développement et des Ressources en Informatique Scientifique of Centre National de la Recherche Scientifique.
Corresponding author address: Dr. Chantal Claud, Laboratoire de Météorologie Dynamique du CNRS, Ecole Polytechnique, 91128 Palaiseau, France.